One high-quality OCTA image per retinal region and gas condition was selected for analysis. High-quality images were defined as those with the fewest artifacts (including motion and segmentation artifacts) and highest signal strength. The superficial and deep retinal layers of the macula and temporal macula were extracted using the commercially available manufacturer software (PlexElite; Carl Zeiss Meditec). The superficial retinal layer included the nerve fiber layer, ganglion cell layer, and the inner nuclear layer. The deep retinal layer included the outer plexiform layer and less than 33% of the outer nuclear layer. Projection artifacts were excluded using the manufacturer software (PlexElite; Carl Zeiss Meditec).
The radial peripapillary capillaries were extracted using the maximum pixel projection customized slab setting with one boundary of the customized slab set to the inner limiting membrane (ILM) and the other displaced 75 µm below the ILM. The superficial 75 µm was used because the radial peripapillary capillaries were most visible and continuous at this depth in our case.29 (link)
The extracted vessel slabs were analyzed using a previously validated software that computed the apparent vessel skeleton density (VSD).30 (link),31 (link) In brief, VSD is a linear measure of vessel length computed as the count of pixels representing vessels (after the OCTA image binarization and skeletonization) divided by the total number of pixels in the image. The software was modified for assessing SS-OCTA images and to coregister images from the different gas conditions. Nonoverlapping regions were excluded from the analysis. In addition, all noncapillary vessels (e.g., large caliber retinal arteries, veins, arterioles and venules) as well as their corresponding negative spaces were excluded from the analysis.18 (link) Anatomic regions known to be characterized by nonperfused OCTA signals (such as foveal avascular zone and disc region) were also excluded (Fig. 2 ).
The choriocapillaris images were also extracted from the macula and temporal macula scans. The Multilayer Segmentation algorithm version 0.7 (PlexElite; Carl Zeiss Meditec) was used to perform choriocapillaris segmentation, which was defined as a 16-µm-thick slab with its anterior boundary located 4 µm beneath Bruch's membrane.32 (link) En face choriocapillaris images were generated using maximum-intensity projection and projection artifact removal.33 (link) Flow deficit density (FDD) and mean flow deficit sizes (MFDSs) were computed for the choriocapillaris images using custom software. The flow deficits (FDs) were segmented from compensated choriocapillaris OCTA en face images as previously described.34 (link) Briefly, pixels were automatically clustered into different groups using the fuzzy C-means approach, and the cluster of pixels with the lowest intensity values was segmented as FDs. With segmented FDs, the choriocapillaris FDD and MFDS (µm2) were calculated for the entire scan. The FDD was defined as the percentage of pixels representing FDs relative to the whole scan, and the MFDS was defined as the average size of all individual FDs in the whole scan.
The radial peripapillary capillaries were extracted using the maximum pixel projection customized slab setting with one boundary of the customized slab set to the inner limiting membrane (ILM) and the other displaced 75 µm below the ILM. The superficial 75 µm was used because the radial peripapillary capillaries were most visible and continuous at this depth in our case.29 (link)
The extracted vessel slabs were analyzed using a previously validated software that computed the apparent vessel skeleton density (VSD).30 (link),31 (link) In brief, VSD is a linear measure of vessel length computed as the count of pixels representing vessels (after the OCTA image binarization and skeletonization) divided by the total number of pixels in the image. The software was modified for assessing SS-OCTA images and to coregister images from the different gas conditions. Nonoverlapping regions were excluded from the analysis. In addition, all noncapillary vessels (e.g., large caliber retinal arteries, veins, arterioles and venules) as well as their corresponding negative spaces were excluded from the analysis.18 (link) Anatomic regions known to be characterized by nonperfused OCTA signals (such as foveal avascular zone and disc region) were also excluded (
The choriocapillaris images were also extracted from the macula and temporal macula scans. The Multilayer Segmentation algorithm version 0.7 (PlexElite; Carl Zeiss Meditec) was used to perform choriocapillaris segmentation, which was defined as a 16-µm-thick slab with its anterior boundary located 4 µm beneath Bruch's membrane.32 (link) En face choriocapillaris images were generated using maximum-intensity projection and projection artifact removal.33 (link) Flow deficit density (FDD) and mean flow deficit sizes (MFDSs) were computed for the choriocapillaris images using custom software. The flow deficits (FDs) were segmented from compensated choriocapillaris OCTA en face images as previously described.34 (link) Briefly, pixels were automatically clustered into different groups using the fuzzy C-means approach, and the cluster of pixels with the lowest intensity values was segmented as FDs. With segmented FDs, the choriocapillaris FDD and MFDS (µm2) were calculated for the entire scan. The FDD was defined as the percentage of pixels representing FDs relative to the whole scan, and the MFDS was defined as the average size of all individual FDs in the whole scan.
